-
Notifications
You must be signed in to change notification settings - Fork 37
2024.01.11 Meeting Notes
Philipp Grete edited this page Jan 25, 2024
·
3 revisions
Agenda
- Individual/group updates
- Potential Gordon Bell submission
- review non-WIP PRs
PM
- Fixed bug for using custom prolong/restriction ops for SparsePools (missed metadata for custom ops). Already merged.
- AMR with curvilinear grids using custom ops in new downstream codes seems to work fine.
- Found issue wrt to which variables are prolongated restricted (we talked about this in the past: cons vs prim) and added a workaround
- FG and PM will touch base and the discuss next time how to go forward (wrt to adding a clean solution or providing a workaround/documentation or ...)
LR
- working downstream (alongside with performance improvements for MG)
JD
- addressed comments in refactored interface to performance instrumentation
- refactored/cleaned up tasking (particularly wrt to iterative tasks)
- builds a graph (rather than while loops with checking for dependencies)
- graph is now persistent rather than being destroyed and recreate every single timestep
- support for nested lists (allows for clean separation of outer/inner solvers/lists)
-
RegionalDependencies
are gone (used for global syncs) and replaced by "task qualifiers" with support for reductions - Task execution is threaded (
ThreadPool
)- BUT GOTCHA rest of Parthenon is not necessarily thread safe (particularly wrt to pack caching)
- might eventually allow for supporting streams again JM
- started initial refactor of IO interface (to eventually allow for faces and edges)
- abstracted/separated non-hdf5 specific, generic ops incl flattening of data
- refactoring/debugging global boundary conditions parsing/enrollment
- but particles boundaries broke (though they'd have to be rewritten soon anyway)
- JM, JD, LR, SM are working on proposal for new downstream code
FG
- curvilinear coordinates for disks works with PLM on static grid
- will sync with PD wrt AMR/prolong/restrict
- working through process to get the code out
JS
- mostly working on physics around numerical relativity
- prolong/restrict for numerical relativity needed special care in AthenaK, e.g., coarse and fine buffers would both have to be communicated (because of higher dimensional operators at higher >4th order)
PG
- actively working on the OpenPMD outputs
- needs some additional interfaces for downstream codes to provide the necessary metadata (such as units)
- currently don't have all the physics implemented that would support a compelling science case that highlights Parthenon capabilities (e.g., inline reaction with sparse variables)
- will keep this in mind (and revisit in ~2 month) so that we have the capabilities in place for 2025